Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add classical filter implementation #1

Open
wants to merge 6 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
132 changes: 132 additions & 0 deletions classical-filter/classical-filter.cpp
Original file line number Diff line number Diff line change
@@ -0,0 +1,132 @@
#include <opencv2/opencv.hpp>
#include <iostream>
#include <cmath>

using namespace cv;
using namespace std;

// Measures how similar two numbers are
double sim(double a, double b) {
return min(a, b) / (a + b);
}

// Calculates the squared Euclidean distance between two points
double distSq(Point2f a, Point2f b) {
return pow(a.x - b.x, 2) + pow(a.y - b.y, 2);
}

// Calculates the angle between two points in degrees
double angle(Point2f a, Point2f b) {
if (a.x == b.x) return 90;
Point2f right = (a.x > b.x) ? a : b;
Point2f left = (a.x > b.x) ? b : a;
return atan((right.y - left.y) / (right.x - left.x)) * 180 / CV_PI;
}

// Applies a mask to detect color-specific contours in the frame
vector<vector<Point>> getContours(Mat& frame, const string& color) {
Mat frameHSV, mask1, mask2, frameThreshold;

cvtColor(frame, frameHSV, COLOR_BGR2HSV);

if (color == "red") {
inRange(frameHSV, Scalar(0, 70, 50), Scalar(20, 255, 255), mask1);
inRange(frameHSV, Scalar(170, 70, 50), Scalar(230, 255, 255), mask2);
} else {
inRange(frameHSV, Scalar(90, 70, 50), Scalar(120, 255, 255), mask1);
inRange(frameHSV, Scalar(170, 70, 50), Scalar(200, 255, 255), mask2);
}
frameThreshold = mask1 | mask2;

erode(frameThreshold, frameThreshold, getStructuringElement(MORPH_ELLIPSE, Size(5, 5)));
dilate(frameThreshold, frameThreshold, getStructuringElement(MORPH_ELLIPSE, Size(5, 5)));
dilate(frameThreshold, frameThreshold, getStructuringElement(MORPH_ELLIPSE, Size(5, 5)));
erode(frameThreshold, frameThreshold, getStructuringElement(MORPH_ELLIPSE, Size(5, 5)));

vector<vector<Point>> contours;
findContours(frameThreshold, contours, RETR_EXTERNAL, CHAIN_APPROX_NONE);

return contours;
}

// Draws the bounding box centers and processes relationships between them
Mat drawCenters(Mat frame, const string& color) {
auto contours = getContours(frame, color);
vector<RotatedRect> bboxes;

for (auto& contour : contours) {
RotatedRect bbox = minAreaRect(contour);
bboxes.push_back(bbox);

Point2f bboxPoints[4];
bbox.points(bboxPoints);
for (int j = 0; j < 4; j++) {
line(frame, bboxPoints[j], bboxPoints[(j + 1) % 4], Scalar(0, 255, 0), 2);
}
}

int thresh = 20;
double widthSimThresh = 0.1, lengthSimThresh = 0.3, yThresh = 15, angleThresh = 15;

for (size_t i = 0; i < bboxes.size(); i++) {
auto& bbox1 = bboxes[i];
Point2f center1 = bbox1.center;
double width1 = bbox1.size.width;
double length1 = bbox1.size.height;
double angle1 = bbox1.angle;

if (max(length1, width1) < thresh) continue;

for (size_t j = i + 1; j < bboxes.size(); j++) {
auto& bbox2 = bboxes[j];
Point2f center2 = bbox2.center;
double width2 = bbox2.size.width;
double length2 = bbox2.size.height;
double angle2 = bbox2.angle;

if (max(length2, width2) < thresh) continue;

double angleDiff = abs(angle1 - angle2);
double yDiff = abs(center1.y - center2.y);

if (sim(width1, width2) > widthSimThresh &&
sim(length1, length2) > lengthSimThresh &&
yDiff < yThresh &&
(angleDiff < angleThresh || angleDiff > 180 - angleThresh)) {
Point centerMid((center1.x + center2.x) / 2, (center1.y + center2.y) / 2);
circle(frame, centerMid, 10, Scalar(255, 0, 255), -1);
}
}
}

return frame;
}

int main() {
VideoCapture cap("tests/test1.mp4");

if (!cap.isOpened()) {
cout << "Error opening video file" << endl;
return -1;
}

while (cap.isOpened()) {
Mat frame;
cap >> frame;

if (frame.empty()) {
cout << "End of video stream." << endl;
break;
}

frame = drawCenters(frame, "blue");

imshow("Frame", frame);

if (waitKey(1) == 'q') break;
}

cap.release();
destroyAllWindows();
return 0;
}
183 changes: 183 additions & 0 deletions classical-filter/classical-filter.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,183 @@
# References:
# https://docs.opencv.org/4.x/dd/d43/tutorial_py_video_display.html
# https://docs.opencv.org/4.x/d9/dc8/tutorial_py_trackbar.html
# https://docs.opencv.org/4.x/d7/d4d/tutorial_py_thresholding.html
# https://www.opencv-srf.com/2010/09/object-detection-using-color-seperation.html
# https://docs.opencv.org/4.5.4/da/d97/tutorial_threshold_inRange.html
# https://docs.opencv.org/4.5.4/db/df6/tutorial_erosion_dilatation.html
# https://docs.opencv.org/3.4/d4/d73/tutorial_py_contours_begin.html
# https://docs.opencv.org/4.x/dc/da5/tutorial_py_drawing_functions.html
# https://stackoverflow.com/questions/38064777/use-waitkey-in-order-pause-and-play-video
# https://docs.opencv.org/4.x/de/d62/tutorial_bounding_rotated_ellipses.html
# https://www.geeksforgeeks.org/python-opencv-cv2-puttext-method/

import cv2 as cv
import numpy as np
import math
import time

color = 'blue'

# Measures how similar two numbers are
def sim(a, b):
return min(a, b) / (a + b)

def dist_sq(a, b):
return (a[0] - b[0]) ** 2 + (a[1] - b[1]) ** 2

def angle(a, b):
if a[0] == b[0]:
return 90
right = a
left = b
if b[0] > a[0]:
right = b
left = a
return math.degrees(math.atan((right[1] - left[1]) / (right[0] - left[0])))

def get_contours(frame, color):
# Apply a red mask to image, apply morphological opening/closing, and find contours of contiguous red areas
frame_HSV = None

mask1, mask2, frame_threshold = None, None, None
if color == 'red':
frame_HSV = cv.cvtColor(frame, cv.COLOR_BGR2HSV)
mask1 = cv.inRange(frame_HSV, (0, 70, 50), (20, 255, 255))
mask2 = cv.inRange(frame_HSV, (170, 70, 50), (230, 255, 255))
frame_threshold = mask1 | mask2
else:
frame_HSV = cv.cvtColor(frame, cv.COLOR_BGR2HSV)
mask1 = cv.inRange(frame_HSV, (90, 100, 100), (115, 255, 255))
mask2 = cv.inRange(frame_HSV, (115, 100, 100), (135, 255, 255))
frame_threshold = mask1 | mask2

frame_threshold = cv.erode(frame_threshold, cv.getStructuringElement(cv.MORPH_ELLIPSE, (5, 5)))
frame_threshold = cv.dilate(frame_threshold, cv.getStructuringElement(cv.MORPH_ELLIPSE, (5, 5)))

frame_threshold = cv.dilate(frame_threshold, cv.getStructuringElement(cv.MORPH_ELLIPSE, (5, 5)))
frame_threshold = cv.erode(frame_threshold, cv.getStructuringElement(cv.MORPH_ELLIPSE, (5, 5)))

contours, _ = cv.findContours(frame_threshold, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_NONE)

return contours

def draw_centers(frame, color):
contours = get_contours(frame, color)

# Compute rotated bounding box for each contour and store in `bboxes`
bboxes = []
for contour in contours:
bbox = cv.minAreaRect(contour)
bboxes.append(bbox)
bbox_points = cv.boxPoints(bbox)
bbox_points = np.intp(bbox_points)
frame = cv.drawContours(frame, [bbox_points], -1, (0, 255, 0), 2)

thresh = 20
width_sim_thresh, length_sim_thresh, y_thresh, angle_thresh = 0.1, 0.3, 0.15, 15
for i in range(len(bboxes)):
bbox1 = bboxes[i]
center1 = (bbox1[0][0], bbox1[0][1])
width1, length1, angle1 = bbox1[1][0], bbox1[1][1], bbox1[2]

vert1 = cv.boxPoints(bbox1)

longest = (0, 1)
bl = vert1[0]
tl = vert1[1]
br = vert1[3]
width1 = math.sqrt(dist_sq(vert1[0], vert1[3]))
if dist_sq(bl, br) > dist_sq(bl, tl):
longest = (0, 3)
width1 = math.sqrt(dist_sq(vert1[0], vert1[1]))
length1 = math.sqrt(dist_sq(vert1[longest[0]], vert1[longest[1]]))
angle1 = angle(vert1[longest[0]], vert1[longest[1]])

# If the bounding box is too small, skip
if max(length1, width1) < thresh:
continue

# how the heck is `minAreaRect` defining the angle
# if angle1 > 45:
# angle1 = 90 - angle1
# width1, length1 = length1, width1

# Matching longer sides is more important, and thus needs a stricter threshold
# if width1 > length1:
# width_sim_thresh, length_sim_thresh = length_sim_thresh, width_sim_thresh

for j in range(i + 1, len(bboxes)):
bbox2 = bboxes[j]
center2 = (bbox2[0][0], bbox2[0][1])
width2, length2, angle2 = bbox2[1][0], bbox2[1][1], bbox2[2]

if max(length2, width2) < thresh:
continue

vert2 = cv.boxPoints(bbox2)

longest = (0, 1)
bl = vert2[0]
tl = vert2[1]
br = vert2[3]
width2 = math.sqrt(dist_sq(vert2[0], vert2[3]))
if dist_sq(bl, br) > dist_sq(bl, tl):
longest = (0, 3)
width2 = math.sqrt(dist_sq(vert2[0], vert2[1]))
length2 = math.sqrt(dist_sq(vert2[longest[0]], vert2[longest[1]]))
angle2 = angle(vert2[longest[0]], vert2[longest[1]])

# if angle2 > 45:
# angle2 = 90 - angle2
# width2, length2 = length2, width2

angle_diff = abs(angle1 - angle2)
y_diff = abs(center1[1] - center2[1])

# If two bounding boxes are similar in size and orientation, place a dot between them
if width_sim_thresh < sim(width1, width2) \
and length_sim_thresh < sim(length1, length2) \
and y_diff < y_thresh * (length1 + length2) / 2 \
and (angle_diff < angle_thresh or angle_diff > 180 - angle_thresh):
cv.circle(frame, (round((bbox1[0][0] + bbox2[0][0]) / 2), round((bbox1[0][1] + bbox2[0][1]) / 2)), 10, (255, 0, 255), -1)
cv.putText(frame, f'w: {sim(width1, width2):.2f}, l: {sim(length1, length2):.2f}, a1: {angle1:.2f}, a2: {angle2:.2f}, {angle_diff:.2f}', (round((bbox1[0][0] + bbox2[0][0]) / 2), round((bbox1[0][1] + bbox2[0][1]) / 2)), cv.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 1, cv.LINE_AA)

# Debugging
# print(f'w --- {i}: {width1}, {j}: {width2}, sim: {sim(width1, width2)}')
# print(f'l --- {i}: {length1}, {j}: {length2}, sim: {sim(length1, length2)}')
# print(f'a --- {i}: {angle1}, {j}: {angle2}, diff: {angle_diff}')

# Reference square to see size of `thresh`
# cv.rectangle(frame, (50, 50), (50 + thresh, 50 + thresh), (255, 0, 255), 1)

return frame

def main():
cap = cv.VideoCapture('tests/test1.mp4')

while cap.isOpened():
# start_time = time.time()

ret, frame = cap.read()

if not ret:
print('Failed to read frame. Exiting...')
break

frame = draw_centers(frame, 'blue')

cv.imshow('frame', frame)

# writer.write(frame)
# cv.imwrite(output, frame)

# end_time = time.time()
# print(end_time - start_time)

if cv.waitKey(50) == ord('q'):
break

cap.release()
cv.destroyAllWindows()

main()